Bitdive Skills

Skills MCP Install Format

Public BitDive workflows packaged for agent use.

This repository contains reusable BitDive skills for runtime trace analysis, MCP-driven investigation, service instrumentation, regression management, and developer workflows. Each skill is a single SKILL.md written to the open Agent Skills standard, so one source installs into Claude Code (and any other agent that supports the standard) instead of rewriting prompts and runbooks by hand.

Demo

Watch the BitDive demo

Watch the BitDive product demo on YouTube:

• https://www.youtube.com/watch?v=WqtLXcODz8I

What This Repository Is For

Use this repository when you want agent-ready BitDive workflows for tasks such as:

• adding BitDive to a Spring Boot service
• understanding which MCP tool to call and when
• comparing traces to find behavioral drift
• reviewing a PR/branch with before/after runtime evidence
• hunting real runtime issues and rejecting false positives
• building trace coverage across a service's endpoints
• running phased BitDive development workflows
• repairing or refreshing replay-based regression groups
• fixing MCP, service, or Docker connectivity to BitDive

These workflows are meant to make BitDive usage repeatable, faster, and less prompt-dependent.

Repository Layout

One skill, one folder, one SKILL.md. There is a single source of truth per skill — no duplicated per-agent copies.

Path	Purpose
skills/<skill-name>/SKILL.md	The skill: YAML frontmatter (name, description) plus the workflow instructions
skills/<skill-name>/agents/openai.yaml	Optional UI metadata (display name, prompt) for installers that surface it

The agent decides when and how to apply each skill from its description — and is free to use several together or run one in isolation as it sees fit.

Included Workflows

Workflow	Purpose
add-bitdive-spring	Add BitDive producer and optional replay support to a Spring Boot service
bitdive-overview	Choose the right BitDive MCP tool for discovery, inspection, comparison, and updates
bitdive-trace-comparison	Compare traces and locate the exact point of behavior drift
bitdive-pr-review	Review a PR/branch with before/after runtime evidence and a merge verdict
bitdive-issue-finding	Hunt real runtime issues and separate verified bugs from false positives
bitdive-endpoint-coverage	Trigger endpoints safely to build trace coverage and document gaps
bitdive-dev-workflow	Run a phased BitDive development workflow with human checkpoints
bitdive-test-management	Create, wire, repair, refresh, and rebuild replay groups
bitdive-connectivity-setup	Fix MCP/service/Docker connectivity to the intended BitDive backend
bitdive-docker-networking	Connect Dockerized services to cloud or self-hosted BitDive

Installation

Claude Code

Install the full skill set into Claude Code with the Agent Skills CLI:

npx skills add bitDive/bitdive-skills --skill '*' -a claude-code

Global (user-wide) install:

npx skills add bitDive/bitdive-skills --skill '*' -g -a claude-code

This places each skill under .claude/skills/<skill-name>/ (or ~/.claude/skills/ for a global install). Claude discovers them automatically and applies a skill when its description matches the task.

Manual install

If you prefer not to use the CLI, copy the skill folders directly:

git clone https://github.com/bitDive/bitdive-skills.git
mkdir -p /path/to/project/.claude/skills
cp -R bitdive-skills/skills/* /path/to/project/.claude/skills/

Other agents

Because each skill follows the open Agent Skills standard, the same source installs into other supported agents by changing the target, for example:

npx skills add bitDive/bitdive-skills --skill '*' -a cursor
npx skills add bitDive/bitdive-skills --skill '*' -a codex

Recommended Environment

These workflows are most useful when:

• BitDive MCP is configured and reachable
• the target service is already instrumented or is about to be instrumented
• the repository has a real module-scoped test command that can be rerun quickly
• the team wants repeatable runtime-backed workflows rather than one-off prompts

Design Notes

This repository is intentionally text-first and workflow-oriented.

• one SKILL.md per skill is the single source of truth — no per-agent duplicates to keep in sync
• the skills are meant to be portable across repositories
• private project names and local-only paths were removed from the public set
• MCP-centered investigation and replay management are the main focus
• each workflow is intended to reduce prompt ambiguity for common BitDive tasks

Notes

• agents/openai.yaml is optional metadata for installers that show a UI; it is not required for the skill to work
• if you want to publish under another license, replace the root LICENSE file before pushing the repository